Active Hidden Naive Bayes

Abstract

Over the years, many learners that take advantage of the Bayesian theory have been developed and proved to be both efficient and performant in terms of classification predictiveness. Hidden Naive Bayes is no exception since its polynomial complexity makes it a desired base classifier to conduct under Weakly Supervised Learning that, unlikely the Supervised Learning, takes advantage of both Labeled and Unlabeled instances in order to create accurate learning models. In this work, we exploit Hidden Naive Bayes under Active Learning scheme, where human interaction is needed for resolving the more disambiguous cases and integrating its knowledge into the learning loop. We compare the proposed Active Learner against 4 state-of-the-art classifiers under the same learning strategy over 14 binary and multiclass datasets.